Pentagon blacklists Anthropic for refusing to drop safety guardrails; its own contracts list Anthropic as a partner anyway.

…(cont) Up on Capitol Hill, the oversight side of this ledger reads thinner. That admission surfaced ten days after a federal appeals court settled a different fight over AI entirely.

That fight is the Pentagon's. In February, it blacklisted Anthropic, calling the company a "supply chain risk," after Anthropic refused to strip the safety guardrails keeping its tools out of mass surveillance and autonomous-weapons work. On August 28th, a judge ruled that blacklist illegal. On September 25th, an appeals court reversed the judge and let the label stand. This week, the BBC confirmed what the paperwork implied all along: the Pentagon has actually stopped using Anthropic's tools. Of everything on this ledger, that's the one entry that cleared.

Here's the gap it clears against. Records The Intercept obtained by lawsuit, published September 8th, show the Pentagon working hand in hand with four companies: OpenAI, Anthropic, Google, and xAI. Anthropic is on both lists — blacklisted for refusing one use of its tools, and still a contractor for everything else it sells the military. This week, CBS also reported that Donald Trump has formed an AI "Super Intelligence Force," a task force charged with planning the future of war. Whose names are on that roster has not made it into the record.

Run the account and the thing running short isn't dollars. It's oversight: 63 percent of the public afraid, a sitting member of Congress a week behind the news he is supposed to regulate, and a Pentagon blacklist narrow enough to bar one company's guardrails while the contracts underneath it keep clearing. The ledger balances, hon. It just doesn't balance for anybody reading it.